Abstract

Given the need for Human-Machine moral Interaction, an AI system that focuses on moral theory involving intersubjectivity is worthy of attention. We will build a network knowledge representation framework based on the second-person standpoint moral theory to analyze the design philosophy of a general artificial intelligence (AGI) moral machine. We then use this framework to investigate an interactive moral machine designed by Bello and Bringsjord (2013). In constructing a feature of mindreading, their study faces philosophical challenges to combine the empirical source configuration of its inheritance function and the multi-world setting of the Polyscheme itself. We try to prove that the integrated structure of the Polyscheme based on a third-person perspective, multi-world setting cannot afford a continuous self that can accommodate enough diversity of the stimulation relationship between the concept of self and other. Moreover, we suggest that the psychological distance structure of mindreading should mainly come from the internal “self” of the system. This requires the system to have an intrinsic kernel, meaning a design philosophy completely different from the integrated style of Polyscheme.